[LEDE-DEV] flash issue

[e9hack]

Hi,

I've some trouble to update my router. It is a TP-Link WDR3600 with an increased flash
(16MB instead of 8MB). If I update the router, I get very often a message after the first
reboot like this:

[   62.840041] jffs2: Newly-erased block contained word 0x19852003 at offset 0x00000000

If an image produces this error, it occurs at every update with this image. It is for
different images the same word and offset. If it occurs, the router works perfectly. WLAN
is up and I don't see any issue. After the next reboot, the router does overwrite most (or
all) files, which are usually generated if sysupgrade is used with option -n. All other
files will be not overwritten. For example, /etc/shadow isn't overwritten. I've some old
configuration files and backups in /etc/config. They will be not overwritten. The old
password is still active and dropbear is running. During my tests, I did restore the
configuration by an previously saved backup. This means, the overlay fs contains all the
time the same, if I update the router with an image.

The message comes from the second entry of this message in linux-4.4.15/fs/jffs2/read.c. I
did modify the code in a way, that at maximum 16 wrong locations will be print. I see 8:

[   61.770067] jffs2: Newly-erased block contained word 0x19852003 at offset 0x00000000
[   61.778274] jffs2: Newly-erased block contained word 0xc at offset 0x00000004
[   61.785816] jffs2: Newly-erased block contained word 0xf060dc98 at offset 0x00000008
[   61.793985] jffs2: Newly-erased block contained word 0x1985e001 at offset 0x0000000c
[   61.802113] jffs2: Newly-erased block contained word 0x36 at offset 0x00000010
[   61.809727] jffs2: Newly-erased block contained word 0x5d4448fe at offset 0x00000014
[   61.817868] jffs2: Newly-erased block contained word 0x1 at offset 0x00000018
[   61.825368] jffs2: Newly-erased block contained word 0x0 at offset 0x0000001c

During implementing this, I generate several images with some small differences. This 8
values are all the time the same. From my experience with i²c eeproms with too much write
cycles, errors starts with single bits in one byte of one page. Eeproms are different than
spi flash’s, but currently I don't assume, that the flash is damaged. I will verify this
by replacing the flash.

I'm using a second router. It is a TP-LINK C7. The cpu is the same family and
configuration to build the image is more or less the same. It has already a 16MB flash by
default. I don't see such errors.

I'm not sure if this error did occur with OpenWrt in the past. I did flash some old
images, but I didn't see this error with OpenWrt images.

Regards,
Hartmut